Order confusion matrix in R

Question

I've created a confusion matrix from the observations and its predictions in 3 classes.

classes=c("Underweight", "Normal", "Overweight")

When I compute the confusion matrix, it organizes the classes in the table alphabetical. Here is my code.

# Confusion matrix
Observations <- bmi_classification(cross.m$bmi)
Predicted <- bmi_classification(cross.m$cvpred)

conf <- table(Predicted, Observations)

library(caret) 
f.conf <- confusionMatrix(conf)
print(f.conf)

This produces this output:

Confusion Matrix and Statistics

             Observations
Predicted     Normal Overweight Underweight
  Normal          17          0           1
  Overweight       1          4           0
  Underweight      1          0           1

So, I would like it to first Underweight, then normal and finally Overweight. I've tried to pass the order to the matrix as an argument but no luck with that.

EDIT:

I tried reordering it,

conf <- table(Predicted, Observations)

reorder = matrix(c(9, 7, 8, 3, 1, 2, 6, 4, 5), nrow=3, ncol=3)

conf.reorder <- conf[reorder]

but I'm getting, [1] 1 1 0 1 17 1 0 0 4

Solution

Try this then redo your code:

 cross.m$Observations <- factor( cross.m$Observations, 
                              levels=c("Underweight","Normal","Overweight") )
 cross.m$Predicted<- factor( cross.m$Predicted, 
                              levels=c("Underweight","Normal","Overweight") )
conf <- table(Predicted, Observations)

library(caret) 
f.conf <- confusionMatrix(conf)
print(f.conf)

Ordinary matrix methods would probably not work since a caret confusion matrix object is a list.

Comments

[priyanka]

This might help

[SamuelNLP]

The bmi_classification function turns a number of bmi into the respective class. Example, if cross.m$bmi = 15 the it returns Underweight, so I don't see how that can help.

[IRTFM]

I editted my response to fit your new code. Although it remains untested in the absence of a reproducible example.